diff --git a/simgear/math/simd.hxx b/simgear/math/simd.hxx
index 81782efa..3074e342 100644
--- a/simgear/math/simd.hxx
+++ b/simgear/math/simd.hxx
@@ -504,10 +504,10 @@ inline simd4_t<float,N>abs(simd4_t<float,N> v) {
 # endif
 
 
-# ifdef __AVX_unsupported__
+# ifdef __AVX__
 #  include <pmmintrin.h>
 #  include <immintrin.h>
-//#  include <avxintrin-emu.h>
+// #  include "avxintrin-emu.h"
 
 template<int N>
 class simd4_t<double,N>
@@ -634,36 +634,29 @@ inline float hsum_pd_avx(__m256d v) {
 
 template<>
 inline double magnitude2(simd4_t<double,4> v) {
-    return hsum_pd_avx(v.v4()*v.v4());
+    return hsum_pd_avx(_mm256_mul_pd(v.v4(),v.v4()));
 }
 
 template<>
 inline double dot(simd4_t<double,4> v1, const simd4_t<double,4>& v2) {
-    return hsum_pd_avx(v1.v4()*v2.v4());
+    return hsum_pd_avx(_mm256_mul_pd(v1.v4(),v2.v4()));
 }
 
+#ifdef __AVX2__
 template<>
 inline simd4_t<double,3> cross(const simd4_t<double,3>& v1, const simd4_t<double,3>& v2)
 {
-#if 1
-    // http://threadlocalmutex.com/?p=8
+    // https://gist.github.com/L2Program/219e07581e69110e7942
     __m256d v41 = v1.v4(), v42 = v2.v4();
-    __m256d a = _mm256_shuffle_pd(v41, v41, _MM_SHUFFLE(3, 0, 2, 1));
-    __m256d b = _mm256_shuffle_pd(v42, v42, _MM_SHUFFLE(3, 0, 2, 1));
-    __m256d c = _mm256_sub_pd(_mm256_mul_pd(v41, b), _mm256_mul_pd(a, v42));
-    return _mm256_shuffle_pd(c, c, _MM_SHUFFLE(3, 0, 2, 1));
-#else
-   v1 = _mm256_sub_pd(
-                 _mm256_mul_pd(
-                    _mm256_shuffle_pd(v1.v4(),v1.v4(),_MM_SHUFFLE(3, 0, 2, 1)),
-                    _mm256_shuffle_pd(v2.v4(),v2.v4(),_MM_SHUFFLE(3, 1, 0, 2))),
-                 _mm256_mul_pd(
-                    _mm256_shuffle_pd(v1.v4(),v1.v4(),_MM_SHUFFLE(3, 1, 0, 2)),
-                    _mm256_shuffle_pd(v2.v4(),v2.v4(),_MM_SHUFFLE(3, 0, 2, 1)))
-   );
-   return v1;
-#endif
+    return _mm256_sub_pd(
+               _mm256_mul_pd(
+                     _mm256_permute4x64_pd(v41,_MM_SHUFFLE(3, 0, 2, 1)),
+                     _mm256_permute4x64_pd(v42,_MM_SHUFFLE(3, 1, 0, 2))),
+               _mm256_mul_pd(
+                     _mm256_permute4x64_pd(v41,_MM_SHUFFLE(3, 1, 0, 2)),
+                     _mm256_permute4x64_pd(v42,_MM_SHUFFLE(3, 0, 2, 1))));
 }
+#endif
 
 template<int N>
 inline simd4_t<double,N> min(simd4_t<double,N> v1, const simd4_t<double,N>& v2) {
diff --git a/simgear/math/simd4x4.hxx b/simgear/math/simd4x4.hxx
index a199d5bb..bbf6908b 100644
--- a/simgear/math/simd4x4.hxx
+++ b/simgear/math/simd4x4.hxx
@@ -514,10 +514,10 @@ inline simd4_t<float,3> transform<float>(const simd4x4_t<float,4>& m, const simd
 # endif
 
 
-# ifdef __AVX_unsupported__
+# ifdef __AVX__
 #  include <pmmintrin.h>
 #  include <immintrin.h>
-//#  include <avxintrin-emu.h>
+// #  include "avxintrin-emu.h"
 
 template<>
 class simd4x4_t<double,4>
@@ -690,25 +690,18 @@ inline simd4x4_t<double,4> rotation_matrix<double>(double angle, const simd4_t<d
 
 template<>
 inline simd4x4_t<double,4> transpose<double>(simd4x4_t<double,4> m) {
-    simd4x4_t<double,4> mtx;
-#if 1
-    for (int i=0; i<4; ++i) {
-        for(int j=0; j<4; ++j) {
-            mtx.ptr()[j][i] = m.ptr()[i][j];
-        }
-    }
-#else
-    __m256d T0 = _mm256_unpacklo_pd(m.m4x4()[0], m.m4x4()[1]);
-    __m256d T1 = _mm256_unpacklo_pd(m.m4x4()[2], m.m4x4()[3]);
-    __m256d T2 = _mm256_unpackhi_pd(m.m4x4()[0], m.m4x4()[1]);
-    __m256d T3 = _mm256_unpackhi_pd(m.m4x4()[2], m.m4x4()[3]);
+// http://stackoverflow.com/questions/36167517/m256d-transpose4-equivalent
+    __m256d tmp0 = _mm256_shuffle_pd(m.m4x4()[0], m.m4x4()[1], 0x0);
+    __m256d tmp2 = _mm256_shuffle_pd(m.m4x4()[0], m.m4x4()[1], 0xF);
+    __m256d tmp1 = _mm256_shuffle_pd(m.m4x4()[2], m.m4x4()[3], 0x0);
+    __m256d tmp3 = _mm256_shuffle_pd(m.m4x4()[2], m.m4x4()[3], 0xF);
 
-    mtx.m4x4()[0] = _mm256_unpacklo_pd(T0, T1);
-    mtx.m4x4()[1] = _mm256_unpackhi_pd(T0, T1);
-    mtx.m4x4()[2] = _mm256_unpacklo_pd(T2, T3);
-    mtx.m4x4()[3] = _mm256_unpackhi_pd(T2, T3);
-#endif
-    return mtx;
+    m.m4x4()[0] = _mm256_permute2f128_pd(tmp0, tmp1, 0x20);
+    m.m4x4()[1] = _mm256_permute2f128_pd(tmp2, tmp3, 0x20);
+    m.m4x4()[2] = _mm256_permute2f128_pd(tmp0, tmp1, 0x31);
+    m.m4x4()[3] = _mm256_permute2f128_pd(tmp2, tmp3, 0x31);
+
+    return m;
 }
 
 inline void translate(simd4x4_t<double,4>& m, const simd4_t<double,3>& dist) {